Nice graphs... in practice, the problem with vignetting calibration isn't so much the fitting to the input images, but rather that sometimes (more so with wider lenses in my experience) the diffuser-on-lens images don't create an accurate model of real-world vignetting behavior. Were you able to check the results on some evenly-lit real-world images for possible over-correction, by any chance?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Easier said than done. In my neighborhood I have found fresh grey mineral plaster on one small building which seemed to be perfect candidate to check the result. The elevation is quite uniform and was well lid (northen side and today there was a fog).
I shot it on 1.5m distance with focus on the wall. There vignietting correction visually are correct. I'm unable to find overcorrections. L variations between corners are not greater than 5 (but in uncorrected image they also occur so the wall was not evenly lid after all).
However...
PanaLeicas are rather known for theirs variations in image quality (see: Lens rental on 25mm or LensTip on 15mm vignetting or LensTip on 20mm vigetting. My point is that I do not know my starting position.
I will shoot more RAW+JPG images (with vignetting corrected in body) and compare the results.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, it's surprisingly difficult to find large, evenly illuminated featureless surfaces... but for checking the corrections you just need a shot that would show if there's visible lightening at the edges of the frame. Clear sky will usually work for this, particularly if the image is viewed at a very small size, which makes it easier to visually notice stuff like that. If a sky shot with the lens set to minimum focus distance doesn't show edge lightening visually, the correction is probably ok.
Unfortunately, for the internally-focusing Micro 4/3 lenses I've tested myself, the vignetting changes substantially between close focus and infinity, with infinity being the worst case. This means profiling at infinity will overcorrect close shots, so it becomes important to either calibrate at multiple distances or just calibrate for the least case, and let infinity still have some vignetting (in-camera correction seems to do something like this). It's a messy situation... thanks for making the effort.
Last edit: junkyardsparkle 2019-02-19
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Your suspicions were correct. At minimum focus distance applied vignetting correction reveals artificially increased luminance of the corners. It doesn't seem right. It's subtle at the beginning but when post-processing the image (increase exposure, adding contrast) it becomes more and more visible.
I can wait for next full overcast sky and do the same procedure with acrylic milk glass but changing the focus distance. So I will set focus to minimum and shoot with full spectrum of aperture. Then repeat this for the next focus distance. Does the change in aperture change the focus point? (not sure about it)
Would this be acceptable? How many focus distance I would need?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Well, considering that the acrylic-on-lens method can also possibly be a source of overcorrection, regardless of focus distance, I would say first try just doing the calibration at minimum focus distance. If applying this correction to shots at infinity produces something similar to in-camera correction, then I would say this is the "safe and simple" approach.
Otherwise, the best distances may depend on the lens. Torsten Bronger's tutorial suggests doubling the MFD once or twice, but some of the lenses we're talking about have very small MFD, so that might not apply as much... it might require some experimenting to see where in the range most of the change takes place. I'll try to do some tests related to this later today.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Oh, and one other important thing is to disable image stabilisation for calibration/testing shots, since it can move the image circle off-center. (Yes, this means that pictures taken with stabilisation are sometimes not perfectly corrected - another reason it may be advisable to err on the side of under-correcting.)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Right. This is the one thing I completly forgot.
One more question. If I intend to do the measurements for various focus distances does it mean that I could stand in one place and change simply the focus distance having image more or less blurred (as for initial measurement focused on infinity)?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Right, so this is the second approach. Unfortunatelly this is done using acylic glass with heavy overcast sky. Can't figure out anything better than that in my environment. I have checked it agaist several pictures and did not observe any overcorrection. My 20mm closeups of flowers look good.
Measurements are for: minimum focus discance - 0.2m , 0.4m, 1.2m and inf. 0.4m is like 33% of scale and 1.2m is near inf (like 75%) for both lenses. Full range of the aperture values are covered.
PDF files with the graphs for both lenses.
Nice graphs... in practice, the problem with vignetting calibration isn't so much the fitting to the input images, but rather that sometimes (more so with wider lenses in my experience) the diffuser-on-lens images don't create an accurate model of real-world vignetting behavior. Were you able to check the results on some evenly-lit real-world images for possible over-correction, by any chance?
Sure, I will. I will shoot it RAW+JPEG and check it. Just give ma a few days.
Last edit: Maciej Patelczyk 2019-02-18
Easier said than done. In my neighborhood I have found fresh grey mineral plaster on one small building which seemed to be perfect candidate to check the result. The elevation is quite uniform and was well lid (northen side and today there was a fog).
I shot it on 1.5m distance with focus on the wall. There vignietting correction visually are correct. I'm unable to find overcorrections. L variations between corners are not greater than 5 (but in uncorrected image they also occur so the wall was not evenly lid after all).
However...
PanaLeicas are rather known for theirs variations in image quality (see: Lens rental on 25mm or LensTip on 15mm vignetting or LensTip on 20mm vigetting. My point is that I do not know my starting position.
I will shoot more RAW+JPG images (with vignetting corrected in body) and compare the results.
Yes, it's surprisingly difficult to find large, evenly illuminated featureless surfaces... but for checking the corrections you just need a shot that would show if there's visible lightening at the edges of the frame. Clear sky will usually work for this, particularly if the image is viewed at a very small size, which makes it easier to visually notice stuff like that. If a sky shot with the lens set to minimum focus distance doesn't show edge lightening visually, the correction is probably ok.
Unfortunately, for the internally-focusing Micro 4/3 lenses I've tested myself, the vignetting changes substantially between close focus and infinity, with infinity being the worst case. This means profiling at infinity will overcorrect close shots, so it becomes important to either calibrate at multiple distances or just calibrate for the least case, and let infinity still have some vignetting (in-camera correction seems to do something like this). It's a messy situation... thanks for making the effort.
Last edit: junkyardsparkle 2019-02-19
Your suspicions were correct. At minimum focus distance applied vignetting correction reveals artificially increased luminance of the corners. It doesn't seem right. It's subtle at the beginning but when post-processing the image (increase exposure, adding contrast) it becomes more and more visible.
I can wait for next full overcast sky and do the same procedure with acrylic milk glass but changing the focus distance. So I will set focus to minimum and shoot with full spectrum of aperture. Then repeat this for the next focus distance. Does the change in aperture change the focus point? (not sure about it)
Would this be acceptable? How many focus distance I would need?
Well, considering that the acrylic-on-lens method can also possibly be a source of overcorrection, regardless of focus distance, I would say first try just doing the calibration at minimum focus distance. If applying this correction to shots at infinity produces something similar to in-camera correction, then I would say this is the "safe and simple" approach.
Otherwise, the best distances may depend on the lens. Torsten Bronger's tutorial suggests doubling the MFD once or twice, but some of the lenses we're talking about have very small MFD, so that might not apply as much... it might require some experimenting to see where in the range most of the change takes place. I'll try to do some tests related to this later today.
Oh, and one other important thing is to disable image stabilisation for calibration/testing shots, since it can move the image circle off-center. (Yes, this means that pictures taken with stabilisation are sometimes not perfectly corrected - another reason it may be advisable to err on the side of under-correcting.)
Right. This is the one thing I completly forgot.
One more question. If I intend to do the measurements for various focus distances does it mean that I could stand in one place and change simply the focus distance having image more or less blurred (as for initial measurement focused on infinity)?
Yes, you don't actually want any "detail" in the calibration images, and don't need it for evaluation images.
Right, so this is the second approach. Unfortunatelly this is done using acylic glass with heavy overcast sky. Can't figure out anything better than that in my environment. I have checked it agaist several pictures and did not observe any overcorrection. My 20mm closeups of flowers look good.
Measurements are for: minimum focus discance - 0.2m , 0.4m, 1.2m and inf. 0.4m is like 33% of scale and 1.2m is near inf (like 75%) for both lenses. Full range of the aperture values are covered.
Thanks for doing the extra work! I'll add the data to the database shortly.